This study employed machine learning that leverage historical experimental data to construct predictive models capable of estimating the oxidation resistance of ultra-high-temperature diboride. The support vector machine regression (SVR) model exhibited superior prediction metrics with R2 of 0.88. The Shapley-Additive-exPlanations and genetic algorithm symbolic regression models revealed that SiC content plays a crucial role in the antioxidant damage of UHT borides, and the results suggested that the critical addition amount of SiC in the ZrB2-SiC system is 13.8 vol%, and the optimal addition amount is 23.5 vol%. The trained SVR model promptly identified optimal formulas exhibiting exceptional oxidation resistance.